Forgejo Repository Datasource

A Dify online document datasource that ingests Forgejo (and Gitea) content
into a Dify Knowledge base — either an overview of the token user's repositories
(README + issues + pull requests), or all text files of one repository's tree
(by file extension), which makes it usable as a real documentation/Knowledge
source rather than just READMEs.

Forgejo is a Gitea fork; both expose the same REST API, so
this plugin works with Gitea too — the shared surface is used for
everything.

Configuration

The plugin needs two credentials:

Credentials are validated on save by calling ; an invalid URL
or token is rejected with a clear message (any embedded in the URL
is redacted from error text).

How to create a Forgejo/Gitea token

1. Sign in to your Forgejo/Gitea server.
2. Go to Settings → Applications → Generate New Token.
3. Give it a name and grant at least read scopes for repositories, issues and
   user (e.g. , , ; the exact scope names
   vary by server version).
4. Copy the generated token and paste it into the Personal Access Token field.

What gets ingested

The datasource has two modes, selected by the Repository parameter.

Repository file mode (set = )

Walks the repository tree and exposes one page per text file (filtered by
extension), so an entire docs/man repo becomes Knowledge:

• File (, or
  on a non-default branch) — the file rendered as Markdown.
• Optionally issues / pull requests (off by default).

Parameters (all optional except where noted):

Default extensions: (see in
). Each file's git blob SHA is stored as the page's
— a precise per-file change marker (note: Dify does not yet
diff online_document pages on it, so re-syncs currently re-ingest all selected
files; the SHA is correct provenance and future-proofs incremental sync).

The tree is read via the recursive git-trees API and paginated, so large
repositories (e.g. a man corpus with tens of thousands of files) are not cut off
at one page.

User mode (leave empty)

For the authenticated user's repositories (from , paginated),
the datasource exposes these pages:

• Repository () — repository metadata plus its README.
• README file () — only when a README exists.
  Gitea/Forgejo has no endpoint, so the plugin lists the repo root
  via and picks a README-named entry.
• Issues () — title, body and comments.
• Pull requests () — title, body and comments.

Each repository page yields a Markdown document; issues and PRs include their
comment threads.

Listing caps

To keep page listings bounded, the datasource caps what it enumerates:

• at most 50 repositories,
• at most 20 issues per repository,
• at most 20 pull requests per repository.

Repositories beyond the cap, and issues/PRs beyond their per-repo caps, are not
listed as pages (the caps are defined in ).

Note that the per-repo issue and pull-request counts are also bounded by the
server's maximum page size: each is fetched in a single request, so a server
whose max page size is below the per-repo cap will return fewer items.

Forgejo/Gitea API endpoints used

• — authenticated user (workspace label, credential check)
• — the user's repositories (paginated)
• — repository metadata (and default branch)
• — repository
  file tree (file mode), paginated
• — repo root listing,
  README detection and file contents
• — issues
• and — issue + comments
• and — pull requests
  (PR comments use the shared endpoint)

All requests authenticate with the header.

Credits

The plugin icon () is the Forgejo logo by
Caesar Schinas, licensed under
CC BY-SA 4.0 and used
unmodified. See . The icon is licensed separately from the
plugin source code (Apache-2.0); per the ShareAlike term it remains under
CC BY-SA 4.0.

License

Apache-2.0. Copyright 2026 Alexey Shabalin.